CN110361000B

CN110361000B - An action event window interception method suitable for motion analysis

Info

Publication number: CN110361000B
Application number: CN201910652069.5A
Authority: CN
Inventors: 王奇; 伊国兴; 缪若琳; 李缘熹; 南熠
Original assignee: Harbin Institute of Technology Shenzhen
Current assignee: Harbin Institute of Technology Shenzhen
Priority date: 2019-07-18
Filing date: 2019-07-18
Publication date: 2021-03-02
Anticipated expiration: 2039-07-18
Also published as: CN110361000A

Abstract

An action event window intercepting method suitable for motion analysis belongs to the technical field of motion analysis. The invention solves the problem of poor accuracy of intercepting the action event window in the existing research method. The method is based on the Teager operator, and additionally adds the energy function of Gaussian smooth filtering, further deduces the modeling process of the action threshold value parameter, determines the self-adaptive threshold value determining scheme taking the energy peak value as the adjusting basis, and adopts the energy function segmentation algorithm based on the threshold value to detect the starting point and the end point of the action, thereby realizing the self-adaptive action signal segment intercepting scheme. The invention can be applied to the technical field of motion analysis.

Description

Action event window intercepting method suitable for motion analysis

Technical Field

The invention belongs to the technical field of motion analysis, and particularly relates to an action event window intercepting method suitable for motion analysis.

Background

The action event window is an intercepting mode aiming at the continuous process of the target action, the action event window can position the initial position of the action and completely cut data under the whole action state, and two ends of the window respectively correspond to the starting state and the ending state of the real action. High-degree-of-freedom motions often show certain complexity, so that an accurate segmentation algorithm is difficult to implement, and meanwhile, a motion event window provides the most complete and clean representation of a target research motion, so that the method is the segmentation mode which is most beneficial to analyzing motion characteristics.

In previous researches, some experimenters realize the intercepting scheme of the action event window by directly setting a threshold value for sensor data, the method is only used for simply judging whether equipment is separated from a static state along with a wearer, and the method has small calculation amount because the data is not required to be additionally processed, but the method has the defects that the intercepting accuracy of the action event window is poor and the method can only be applied to a data curve obtained under a strict laboratory acquisition scene.

Disclosure of Invention

The invention aims to solve the problem that the accuracy of intercepting an action event window is poor in the existing research method.

The technical scheme adopted by the invention for solving the technical problems is as follows: an action event window intercepting method suitable for motion analysis, the method comprises the following steps:

the method comprises the following steps that firstly, data acquisition is carried out on the same action of a human body by using an accelerometer and a gyroscope respectively, and output values of the accelerometer and the gyroscope at each sampling point are obtained;

respectively calculating energy sequences output by the accelerometer and the gyroscope according to output values of the accelerometer and the gyroscope at each sampling point;

step two, smoothing the energy sequences output by the accelerometer and the gyroscope respectively to obtain smooth energy functions output by the accelerometer and the gyroscope;

step three, respectively calculating the self-adaptive energy threshold a of the accelerometer and the gyroscope_thAnd ω_th；

Step four, according to the adaptive energy threshold a of step three_thAnd ω_thDetermining the starting point and the end point of the human body action segment, and completing an action event windowAnd (4) intercepting.

The invention has the beneficial effects that: the invention provides an action event window intercepting method suitable for motion analysis, which is based on a Teager operator and additionally adds an energy function of Gaussian smooth filtering, further deduces the modeling process of an action threshold parameter, determines an adaptive threshold determining scheme taking an energy peak value as an adjusting basis, and detects the starting point and the end point of an action by adopting an energy function segmentation algorithm based on a threshold value, thereby realizing the adaptive action signal section intercepting scheme.

Drawings

FIG. 1 is a flow chart of an action event window capture method suitable for motion analysis according to the present invention;

FIG. 2 is a graph of the effect of clipping data collected by a gyroscope;

FIG. 3 is a graph of the effect of clipping on data collected by an accelerometer;

FIG. 4 is a graph of the effect of using a squared differential energy method to intercept gyroscope data;

FIG. 5 is a diagram of the effect of using the Teager energy method to intercept gyroscope data;

FIG. 6 is a schematic diagram of smoothing an energy sequence under different window length values;

FIG. 7 is an enlarged view of a portion of FIG. 6;

FIG. 8 is a schematic diagram of the corresponding split-bit line using the Teager energy method;

FIG. 9 is a histogram of the corresponding energy distribution of FIG. 8;

FIG. 10 is a quantile value X_rA graph of variation of (d);

FIG. 11 is a quantile value X_rA differential graph of (a);

fig. 12 is a graph of reference threshold versus energy peak.

Detailed Description

The first embodiment is as follows: as shown in fig. 1, the method for intercepting an action event window suitable for motion analysis according to this embodiment includes the following steps:

Step four, according to the adaptive energy threshold a of step three_thAnd ω_thAnd determining the starting point and the end point of the human body action segment, and finishing the interception of the action event window.

The second embodiment is as follows: the first difference between the present embodiment and the specific embodiment is: the expression of the energy sequence output by the accelerometer and the gyroscope in the first step is specifically as follows:

the expression for the energy series output by the accelerometer is:

wherein: e_acc(k-1) represents the output energy value of the accelerometer at the (k-1) th sampling point, i ═ x, y, z, a_x(k) Representing the output value of the accelerometer at the k-th sampling point in the x-axis direction, a_y(k) Representing the output value of the accelerometer at the kth sampling point in the y-axis direction, a_z(k) Representing the output value of the accelerometer at the k sampling point in the z-axis direction; k is 2,3, …, n-1, k represents the kth sampling point, and n represents the number of sampling points;

the x axis, the y axis and the z axis refer to the x axis, the y axis and the z axis under a space rectangular coordinate system.

The expression of the energy sequence output by the gyroscope is:

wherein: e_gyro(k-1) represents the output energy value of the gyroscope at the (k-1) th sampling point, ω_x(k) Represents the output value, ω, of the gyroscope at the k-th sampling point in the x-axis direction_y(k) Represents the output value, ω, of the gyroscope at the kth sampling point in the y-axis direction_z(k) Representing the output value of the gyroscope at the kth sampling point in the z-axis direction.

a_x(k +1) represents the output value of the accelerometer at the (k +1) th sampling point in the x-axis direction, a_y(k +1) represents the output value of the accelerometer at the (k +1) th sampling point in the y-axis direction, a_z(k +1) represents the output value of the accelerometer at the (k +1) th sampling point in the z-axis direction, a_x(k-1) represents the output value of the accelerometer at the k-1 th sampling point in the x-axis direction, a_y(k-1) represents the output value of the accelerometer at the k-1 th sampling point in the y-axis direction, a_z(k-1) represents an output value of the accelerometer at a k-1 th sampling point in the z-axis direction;

ω_x(k +1) represents an output value of the gyroscope at the (k +1) th sampling point in the x-axis direction, ω_y(k +1) represents an output value of the gyroscope at the (k +1) th sampling point in the y-axis direction, ω_z(k +1) represents an output value of the gyroscope at a (k +1) th sampling point in the z-axis direction;

ω_x(k-1) represents the output value of the gyroscope at the k-1 th sampling point in the x-axis direction, ω_y(k-1) represents the output value of the gyroscope at the (k-1) th sampling point in the y-axis direction, ω_z(k-1) represents an output value of the gyroscope at a k-1 th sampling point in the z-axis direction.

The third concrete implementation mode: the second embodiment is different from the first embodiment in that: the specific process of the second step is as follows:

respectively smoothing energy sequences output by the accelerometer and the gyroscope, wherein the smoothing is implemented by Gaussian filtering;

setting the window length of the filter as b, the standard deviation parameter as sigma, and the gaussian kernel vector template as:

the convolution operation is implemented using a sliding window method as follows: performing dot product on the data in the window and a Gaussian kernel vector template to obtain a smooth energy function output by the accelerometer and the gyroscope;

the expression of the smoothed energy function output by the accelerometer and the gyroscope is specifically as follows:

wherein: e_acc(k) Representing the output energy value of the accelerometer at the kth sampling point, y_filted(k) An energy value at a kth sampling point of a smoothed energy function representing an accelerometer output;

E_gyro(k) represents the output energy value, y ', of the gyroscope at the kth sampling point'_filted(k) The energy value of the smoothed energy function representing the gyroscope output at the kth sampling point.

The fourth concrete implementation mode: the third difference between the present embodiment and the specific embodiment is that: the specific process of the third step is as follows:

adaptive energy threshold a for accelerometers_thThe calculation formula of (2) is as follows:

a_th＝0.0267max[E_acc(k)]+19

wherein: e_acc(k) Representing the output energy value of the accelerometer at the kth sampling point;

self-adaptive energy threshold omega of gyroscope_thThe calculation formula of (2) is as follows:

wherein: e_gyro(k) Representing the output energy value of the gyroscope at the kth sampling point.

The fifth concrete implementation mode: the fourth difference between this embodiment and the specific embodiment is that: the specific process of the step four is as follows:

step four, traversing from the starting point of the smooth energy function output by the accelerometer, recording whether the accelerometer is in the action section by using a state switch, and setting the state switch to be 0 in the starting point of the smooth energy function output by the accelerometer in a default mode;

step four, for the energy value y of the smoothed energy function output by the accelerometer at the k sampling point_filted(k) If y is_filted(k)≥a_thIf so, the kth sampling point is considered as a possible starting point of the action state of the accelerometer signal, and the state switch is temporarily set to be 1; in order to avoid the problem of local misjudgment, continuously traversing and investigating whether the energy value of the sampling point after the kth sampling point is greater than a threshold value;

starting from the k sampling point, if the energy values of the smooth energy function output by the accelerometer are all more than or equal to a at the continuous N sampling points_thIf not, determining that the kth sampling point is not the starting point of the action state of the accelerometer signal, and resetting the state switch to 0;

for accelerometer output energy value y at kth sampling point_filted(k) If y is_filted(k)＜a_thDirectly determining that the kth sampling point is not the starting point of the action state of the accelerometer signal;

step three, if the kth sampling point is not the starting point of the action state of the accelerometer signal, repeating the process of the step two, continuously judging the kth +1 th sampling point until the starting point of the action state of the accelerometer signal is determined, and continuously executing the step four;

if the kth sampling point is the starting point of the action state of the accelerometer signal, directly executing the fourth step;

fourthly, in the smooth energy function output by the accelerometer, traversing from the next sampling point of the starting point of the action state of the accelerometer signal until the output energy value y of the accelerometer at the kth sampling point appears_filted(k′)＜a_thDetermining that the k' -1 sampling point is the terminal point of the action state of the accelerometer signal;

step IV, for the smooth energy function output by the gyroscope, repeating the process from the step IV to determine a starting point k of the action state of the gyroscope signal_gyroAnd terminal k'_gyro；

Step four, determining a starting point and an end point of a human body action section according to the starting point and the end point of the accelerometer signal action state and the starting point and the end point of the gyroscope signal action state;

start＝min(k_acc,k_gyro)

termi＝max(k′_acc,k′_gyro)

wherein: start and term represent the start and end points of the body motion segment, k_accAnd k'_accRespectively representing the start and end points of the accelerometer signal, k_gyroAnd k'_gyroRepresenting the start and end points of the gyroscope signal, respectively.

After the fourth step, the fourth step is completed, whether the condition that k' -k is larger than 50 is met or not needs to be judged, if the condition is met, the fourth step, the fifth step and the fourth step, the sixth step are continuously executed, if the condition is not met, the collected action data are short, and the data collected at this time are directly discarded. The accelerometer and the gyroscope are reused for carrying out data acquisition on the same action of the human body, the processes of the first step, the second step, the third step and the fourth step are repeatedly carried out until k' -k is more than 50, the processes are stopped, and the starting point and the end point of the action state of the accelerometer signal obtained at the last time are taken as the starting point k of the action state of the accelerometer signal_accAnd terminal k'_acc；

Repeating the process from step four to the smooth energy function output by the gyroscope obtained at the last time, and determining the starting point of the action state of the gyroscope signalk_gyroAnd terminal k'_gyro. The method can determine the starting point and the end point of the action and complete the interception of the action event window. The final interception effect is shown in fig. 2 and 3:

the sixth specific implementation mode: the fifth embodiment is different from the fifth embodiment in that: the window length b of the filter has a value of 20.

The seventh embodiment: the sixth embodiment is different from the sixth embodiment in that: the value of the standard deviation parameter sigma is 0.627.

The specific implementation mode is eight: the seventh embodiment is different from the seventh embodiment in that: and estimating with the sampling rate of 200Hz, wherein the value of N in the fourth step is 10.

Examples

Function of energy

The output signal is usually relatively stable in the stages before the motion is generated and after the motion is finished, and the signal can be instantaneously and violently changed along with the start of the motion, and the characteristic forms a basic basis for segmenting the motion window.

In real motion capture, it is not possible for the device wearer to maintain a perfect resting level before and after the motion segment, especially in outdoor sports activities such as some instruments, where the body of the exerciser involuntarily shakes to a certain extent by bearing the weight of the equipment during the motion break-out period. The so-called "onset" of motion is not just a simple change from nothing to nothing, but is a relative concept influenced by previous motion levels, and therefore it is necessary to construct a more reasonable representation of the motion level, i.e. an energy function of the motion.

Accelerometers are not zero output level at rest, thus naturally envisaging the design basis for using differential signals as a function of energy. Meanwhile, in order to amplify the fluctuation degree of the signal, the differential value is squared to obtain a basic scheme for constructing an energy function, and x is set_t(τ) represents a single-dimensional discrete sequence, and the square difference energy function operator is expressed as formula (1):

let a (k), k is 1,2, …, n represents the MEMS accelerometer output value at the k-th sampling point, and ω (k) represents the MEMS gyroscope output value at the k-th sampling point. Defining an energy function of the data of the triaxial sensor as the sum of energy values on three coordinate axes, and applying an operator in the formula (1) to obtain calculation formulas of linear motion energy and angular motion energy as shown in formulas (2) and (3) respectively:

in addition, in the field of signal processing, there is an Energy Operator called a Teager Energy Operator (TEO), which is shown in formula (4):

substituting into accelerometer and gyroscope data sequence, namely substituting into formula (2) and formula (3), obtain the computational formula of second linear motion energy and angular motion energy and show as formula (5), (6) respectively:

in order to select the most suitable energy function from the two schemes, two operators are respectively adopted to calculate the angular motion energy function of a piece of gyroscope data of the most basic action, and the comparison effect is shown in fig. 4 and fig. 5:

it can be obviously seen from the comparative experiment effect diagram that the Teager energy function obviously shows better action segment distinguishing characteristics, can inhibit the swing amplitude of the signal to a certain extent, and shows a certain self-adjusting capability when tracking the signal energy, which is the reason why the TEO operator can be widely applied in many signal research fields. The invention finally determines the index function taking the Teager energy as the action degree.

Smoothing function

The availability of the Teager energy function was demonstrated previously, but in fact the TEO operator can only suppress environmental disturbances and random jitter of the wearer to a limited extent. For more complex motions, it is also inevitable that using the Teager's energy function will violate the expected curve shape, making it difficult to cleanly segment the motion segment. The solution is to smooth the energy value curve.

The essence of the smoothing function is filtering, which operates on the data in windows by reducing the noise content of the data to reduce the appearance of detail (spike) in the curve, usually by linear operation, by "swiping" the curve through a small window of constant width.

The adopted smoothing method is Gaussian filtering, and the method has the advantage that the distribution characteristics can be more reserved while smoothing. The realization mode is sliding window convolution, and the value of each smoothed data point is obtained by the Gaussian kernel convolution of the data point and other data points in the neighborhood. The one-dimensional gaussian kernel function is actually a normally distributed density function, and the expression thereof is shown in formula (7):

the image of the Gaussian function is a bell-shaped curve, the closer to the center, the larger the value, and the farther away from the center, the smaller the value. A filter using a gaussian kernel is a low-pass filter capable of attenuating high-frequency signals. Before the smoothing operation is performed, firstly, a numerical gaussian kernel vector template needs to be calculated: assuming that the window length of the filter is b, after the value of the parameter σ is determined, a determined real number sequence with a length of 2b-1 can be obtained, and the calculation method is shown in formula (8):

in order to calculate the smoothing value at a point y (i) on the sequence to be smoothed, only the ith data point is taken as the center, b-1 data segments before and after the ith data point are taken as a smoothing neighborhood and a Gaussian kernel vector template to perform dot product calculation, namely, the formula (9) is shown as follows:

to determine the appropriate value of the parameter σ, consider a gaussian filter with a standard deviation parameter σ, assuming that the standard deviation of the original data is of size σ_XStandard deviation σ of the processed data_rWill decay to about as shown in equation (10):

then, in the actual processing, the energy of the smoothed data is attenuated to 40%, and the estimation is performed by substituting equation (10), so that σ ≈ 0.627 is obtained.

Next, a gaussian smoothing experiment is performed on the Teager energy signal of a section of action, and the gaussian smoothing effect under different values is observed as shown in fig. 6, and fig. 7 is a partial enlarged view of fig. 6.

From the partial enlarged view around the action starting point, with the energy value at the known starting point as the threshold, some data points after the starting point may still have the phenomenon that the instantaneous energy is lower than the threshold, and they will be determined as non-action segment data, thus constituting the phenomenon of misjudgment domain in threshold segmentation. A suitable smoothing process can eliminate this effect because the false positives of these short term fluctuations increase by a weighted average under the trend of the ambient energy values rising, no longer being below the threshold point. Too narrow a smooth window can lead to excessive data details to be retained, insufficient processing strength still causes a false domain-judging phenomenon, and if the window is too large, the data curve can be seriously distorted, the reliability of the threshold value is reduced, and finally, the appropriate window length is determined to be 20 points through multiple groups of experiments.

Threshold determination

The determination of the threshold parameter is the core of the truncation algorithm. Due to the diversity of the motion modes covered in the set, the energy base number of each motion is different, and even the amplitude difference between different executors under the same type of motion is quite large, so that it is not practical to use a constant value as a threshold value to complete the interception of all the motions. There is a need to propose an adaptive threshold determination scheme. First, a graph of gyroscope data energy for a sample motion is shown in FIG. 8.

Quantile is a statistical concept, P (X < X)_r) When r is equal to X_rIs a sequence [ x]R (%) quantile value of (c);

from fig. 8, a series of sets of place values are calculated for the energy sequence (after smoothing), and the place lines at different percentiles are plotted, it can be seen that the distribution of a segment of motion energy values is mainly concentrated in the motion stationary segment, because the motion signal is mainly caused by the random body jitter of the wearer, and the fluctuation degree of the data points is high. This behavior is more apparent if the energy distribution histogram is plotted with the energy value interval as the argument as shown in fig. 9:

and modeling the action threshold by utilizing the dense distribution characteristic of the energy values in the low-amplitude motion segment. From fig. 9, from the low decimals to the high decimals, the distribution of the energy decimals is gradually sparse, that is, as the percentile increases, the increase of the decimals gradually increases, so an experimental assumption is made as to the change rule of the decimals: at the beginning of the action segment, the quantile value will increase dramatically.

To confirm this hypothesis, the variation curves of the fractional values under uniform amplification of r values from 1% to 70% and their difference curves were plotted, as shown in fig. 10 and 11, respectively:

it can be observed from FIGS. 10 and 11 that the value is near 5A steep trend occurs at a quantile value of 0%. The place value is substituted into an inertia data curve graph for verification, the threshold value of the place value is basically consistent with the threshold value of the starting point and the end point of the data, and the rule is basically verified on the data of other types of actions. The particular quantiles obtained in the entire hypothesis-flow experiment are then referred to as the maximum cluster quantile value

The significance is that the method embodies the concentration effect of the low-amplitude motion segment to the maximum extent, the quantiles are continuously improved, and the distance between the quantiles is obviously increased. The mathematical determination method for maximizing the clustering quantile value not only needs to draw a quantile value change curve first, but also needs to perform polyline fitting by taking the minimized square difference as an objective function and to take the inflection point of the polyline fitting, and the calculation process is undoubtedly very complex in practical application.

In order to simplify the model, the maximum cluster quantile value is used as a reference threshold value, and a scatter diagram of the relationship between the maximum cluster quantile value and the energy peak value is drawn on an experimental data set as shown in FIG. 12, so that a clear linear relationship can be found between the maximum cluster quantile value and the energy peak value. Also targeted at minimizing the squared error is the best fit polynomial as shown in equation (11):

the reference threshold is a linear function related to the peak energy, and the experimental result is also consistent with the above description: i.e. the smoothness of the non-motion segment is an expression of the degree of motion "relative" the energy peaks will influence the specific value of the threshold. The slope in the above equation is retained as an adjustment parameter, and the determined threshold calculation template is as shown in equation (12): :

the threshold base is slightly changed from equation (11) because, according to experiments, better segmentation results are achieved overall by taking this value.

The above experimental processes are all based on gyroscope data, and the energy threshold of the accelerometer data is determined by similar experimental steps as shown in formula (13):

a_th＝0.0267max[E_acc(k)]+19 (13)

the above-described calculation examples of the present invention are merely to explain the calculation model and the calculation flow of the present invention in detail, and are not intended to limit the embodiments of the present invention. It will be apparent to those skilled in the art that other variations and modifications of the present invention can be made based on the above description, and it is not intended to be exhaustive or to limit the invention to the precise form disclosed, and all such modifications and variations are possible and contemplated as falling within the scope of the invention.

Claims

1. an action event window interception method that is applicable to motion analysis, is characterized in that, this method may further comprise the steps:

Step 1. Use the accelerometer and the gyroscope to collect data on the same action of the human body, and obtain the output value of the accelerometer and the gyroscope at each sampling point;

According to the output value of the accelerometer and the gyroscope at each sampling point, calculate the energy sequence output by the accelerometer and the gyroscope respectively;

The expression of the energy sequence output by the accelerometer and the gyroscope in the step 1 is specifically:

The expression for the energy sequence output by the accelerometer is:

Where: E _acc (k-1) represents the output energy value of the accelerometer at the k-1th sampling point, i=x, y, z, a _x (k) represents the kth sampling of the accelerometer in the x-axis direction The output value at the point, a _y (k) represents the output value of the accelerometer at the k-th sampling point in the y-axis direction, and a _z (k) represents the output value of the accelerometer at the k-th sampling point in the z-axis direction; k=2,3,...,n-1, k represents the kth sampling point, n represents the number of sampling points;

The expression of the energy sequence output by the gyroscope is:

Where: E _gyro (k-1) represents the output energy value of the gyroscope at the k-1th sampling point, ω _x (k) represents the output value of the gyroscope at the kth sampling point in the x-axis direction, ω _y (k) represents the output value of the gyroscope at the k-th sampling point in the y-axis direction, and ω _z (k) represents the output value of the gyroscope at the k-th sampling point in the z-axis direction;

Step 2: Smoothing the energy sequences output by the accelerometer and the gyroscope, respectively, to obtain a smoothed energy function output by the accelerometer and the gyroscope;

Step 3: Calculate the adaptive energy thresholds a _th and ω _th of the accelerometer and the gyroscope respectively;

Step 4: Determine the start point and end point of the human action segment according to the adaptive energy thresholds a _th and ω _th in step 3, and complete the interception of the action event window.

2. a kind of action event window interception method that is applicable to motion analysis according to claim 1, is characterized in that, the concrete process of described step 2 is:

The energy sequences output by the accelerometer and the gyroscope are respectively smoothed, and the smoothing method is Gaussian filtering;

Set the window length of the filter to b and the standard deviation parameter to σ, then the Gaussian kernel vector template is:

Among them: G-kernel(j) represents the Gaussian kernel vector template;

Do the dot product between the data in the window and the Gaussian kernel vector template to obtain the smooth energy function output by the accelerometer and gyroscope;

The expression of the smoothed energy function output by the accelerometer and gyroscope is specifically:

Where: E _acc (k) represents the output energy value of the accelerometer at the k-th sampling point, y _filtered (k) represents the energy value of the smoothed energy function output by the accelerometer at the k-th sampling point;

E _gyro (k) represents the output energy value of the gyroscope at the kth sampling point, and y′ _filtered (k) represents the energy value of the smoothed energy function output by the gyroscope at the kth sampling point.

3. a kind of action event window interception method that is applicable to motion analysis according to claim 2, is characterized in that, the concrete process of described step 3 is:

The formula for calculating the adaptive energy threshold a _th of the accelerometer is:

a _th =0.0267max[E _acc (k)]+19

Where: E _acc (k) represents the output energy value of the accelerometer at the k-th sampling point;

The formula for calculating the adaptive energy threshold ω _th of the gyroscope is:

Where: E _gyro (k) represents the output energy value of the gyroscope at the kth sampling point.

4. a kind of action event window interception method applicable to motion analysis according to claim 3, is characterized in that, the concrete process of described step 4 is:

Step 41: Start traversing from the starting point of the smooth energy function output by the accelerometer, and use the state switch to record whether it is in the action segment. The state switch is set to 0 by default at the starting point of the smooth energy function output by the accelerometer;

Step 42: For the energy value y _filted (k) of the smooth energy function output by the accelerometer at the k th sampling point, if y _filted (k) ≥ a _th , the k th sampling point is considered to be the accelerometer signal action The possible starting point of the state, temporarily set the state switch to 1;

Starting from the kth sampling point, if the energy values of the smooth energy function output by the accelerometer are all greater than or equal to a _th at consecutive N sampling points, then the kth sampling point is determined to be the starting point of the accelerometer signal action state, Otherwise, determine that the kth sampling point is not the starting point of the action state of the accelerometer signal, and reset the state switch to 0;

For the accelerometer output energy value y _filtered (k) at the k th sampling point, if y _filtered (k)<a _th , it is directly determined that the k th sampling point is not the starting point of the accelerometer signal action state;

Step 43: If the kth sampling point is not the starting point of the action state of the accelerometer signal, repeat the process of step 42, and continue to judge the k+1th sampling point until the starting point of the action state of the accelerometer signal is determined. , continue to step 4 and 4;

If the kth sampling point is the starting point of the action state of the accelerometer signal, step 4 and 4 are directly executed;

Step 44. In the smooth energy function output by the accelerometer, traverse from the next sampling point at the starting point of the accelerometer signal action state until the accelerometer output energy value y _filtered (k') at the k'th sampling point appears <a _th , determine that the k′-1th sampling point is the end point of the accelerometer signal action state;

Step 45: For the smooth energy function output by the gyroscope, repeat the process of steps 41 to 44 to determine the starting point k _gyro and the end point k' _gyro of the gyroscope signal action state;

Step 46: According to the starting point and the ending point of the action state of the accelerometer signal and the starting point and the end point of the action state of the gyroscope signal, determine the starting point and the end point of the human body action segment;

start=min(k _acc ,k _gyro )

termi=max(k′ _acc ,k′ _gyro )

Among them: start and termi represent the start and end points of the human action segment, respectively, and k _acc and k′ _acc represent the start and end points of the accelerometer signal, respectively.

5 . The method for intercepting an action event window suitable for motion analysis according to claim 4 , wherein the value of the window length b of the filter is 20. 6 .

6 . The action event window interception method suitable for motion analysis according to claim 5 , wherein the value of the standard deviation parameter σ is 0.627. 7 .

7 . The method for intercepting an action event window suitable for motion analysis according to claim 6 , wherein the value of N in the step 42 is 10. 8 .